Lovable security is back in the news. On November 14, 2026, a study by the Hamburg security firm Crissy Field reached the front page of Hacker News: of 159,041 Lovable projects it found in the public web archive, thousands of apps returned personal data to anyone who asked, with no login. The cause was not a hack of Lovable or Supabase. It was missing or loose database access rules inside the apps themselves. This post covers what the study found, what Lovable says, how to check your own app today, and what it changes if you build apps with AI.

Quick Answer
- What happened: Crissy Field, working with the German daily Handelsblatt, analyzed 159,041 Lovable projects online between December 2025 and September 2026. 34,545 used Supabase as their database.
- What was exposed: 24,664 of those sites sent database records on page load. With further read-only requests using the same public key, the researchers found tables with email addresses on 7,403 sites (30% of that group), plus passport details, ticket and payment records and more than 200,000 chat messages in individual cases.
- Why: Supabase apps ship a public key to the browser by design. The data is only protected by Row Level Security (RLS) rules in the database. When those rules are missing or too permissive, anyone can read the tables.
- Lovable's position: it calls this a known, industry-wide Supabase configuration problem, and points to its built-in security scans.
- What to do now: run Lovable's Deep scan, review RLS on every table, and test your app logged out before it holds real customer data.
What happened: the November 2026 Lovable security study
The research comes from Crissy Field GmbH, an application security consultancy in Hamburg. Handelsblatt reporter Lina Knees first published the findings in German on November 2, 2026. Crissy Field then released an English report with updated figures, which hit the Hacker News front page on November 14, 2026 with more than 780 points and 300 comments. Coverage followed the same day in Trending Topics and other outlets.
The method was simple, which is the point. The team used Common Crawl, the open web archive, to find Lovable projects. It focused on the ones connected to Supabase, because that integration is easy to analyze and, in the researchers' view, the most common database setup among Lovable apps. Lovable's own database features, introduced in late 2025, were not examined.
For each site, they looked at two things: the data a page sends when it first loads, and what else came back from extra read-only requests made with the same API key the page already exposes.
The numbers
| Step | Count | What it means |
|---|---|---|
| Lovable projects found in Common Crawl (Dec 2025 to Sep 2026) | 159,041 | The sample, not all Lovable apps |
| Projects using Supabase | 34,545 | The group the study analyzed |
| Sites sending database records on first load, no login | 24,664 | Not a problem by itself; some data is meant to be public |
| Sites where further reads returned tables with email addresses | 7,403 (30% of 24,664) | 10,560 tables in total |
Source: Crissy Field, "Lovable: Many Apps Built on the Platform Expose Sensitive Data", November 2026.
The researchers are careful about that 30%. Email addresses can be public contact details or test data, and the method misses personal records with no email field. They say it should not be read as a confirmed leak rate, and that a Common Crawl sample is not representative of every Lovable app.
What was actually reachable
The individual cases are what made the story spread:
- A German lifestyle forum exposed more than 80,000 profiles (names, emails, dates of birth, gender) and more than 200,000 chat messages.
- Event organizers and restaurants exposed tens of thousands of ticket and food-order records with emails, phone numbers, addresses, payment information and ticket codes.
- A property manager exposed AI-generated summaries of internal emails about tenants and owners.
- One service provider exposed more than 1,000 identity document records, including passport numbers, expiry dates and nationalities.
- In another case, 280,000 emails sent to a provider's customers were readable.
Crissy Field did not publish the affected apps or the exact access method, and it could not tell whether any of the data had been misused.
Why this keeps happening with Supabase-backed AI apps
Nobody broke into Lovable or Supabase. The study describes a configuration pattern, and it helps to understand it in one paragraph.
A Supabase app talks to its database directly from the browser. To do that, the page carries a public "anon" key. That key is public by design, so the real protection is Row Level Security: Postgres rules that decide which rows each request may read or change. If a table has RLS off, or a policy that effectively says "everyone may read", the public key is enough to pull the table.
Two things make this worse for AI-built apps:
- The patterns repeat. Crissy Field notes that Lovable's database integrations follow recurring patterns, so a check that works on one app works on thousands with little extra effort.
- The builder is often not a developer. Lovable markets that no technical knowledge is needed, while its terms put responsibility for security and data handling on the customer. People without a security background have a hard time spotting a missing policy.
It is also not unique to Lovable. Trending Topics reports that UpGuard examined about 300,000 Supabase-linked domains and found readable tables on 16,326 databases. Bolt.new and v0 can also connect to Supabase.
What Lovable says, and what it already offers
According to Cybernews, as cited by Trending Topics, Lovable does not see a flaw in its AI and describes the issue as a known, industry-wide Supabase configuration problem.
Lovable does ship real tooling here. Its security documentation (checked November 15, 2026) describes:
- Quick scan, which runs on every publish and flags tables without row-level security, access rules that let everyone through, leaked-password protection turned off, vulnerable npm dependencies and MCP servers exposed without authentication.
- Deep scan, which reviews the app's code for access control and authorization issues specific to your logic.
- Optional Wiz scanning and AI penetration testing through Aikido, and scheduled scans on Enterprise.
The same page states that these tools "cannot guarantee complete security" and that you are responsible for your app's security. That matches the study's main point: a scan on publish only helps if the owner reads the findings and fixes them.
This is Lovable's second security headline of 2026. In April, Lovable published a post-mortem after a researcher showed that chat history and source code of public projects were readable by any logged-in user between February 3 and April 20, 2026. That was a platform bug, fixed within two hours. The November study is different: it is about apps whose owners never locked their data down. If you want the broader picture of the product, our Lovable review covers pricing, strengths and limits.
How to check your Lovable app today
If you run a Lovable app with real users, do this before anything else. It takes under an hour for a small app.
- Run the Deep scan from the project's Security view and fix every database and access-control finding first.
- List every table in Supabase and confirm RLS is enabled on each one. A table with RLS off is readable with the public key.
- Read each policy. Policies like
using (true)on a table with personal data mean anyone can read it. Scope reads toauth.uid()or to the rows a user owns. - Test logged out. Open your app in a private window, open the network tab, and look at what comes back. If you see other people's data, so can anyone else.
- Check tables you forgot. Waitlists, contact forms, newsletter signups and logs held many of the exposed records in the study.
- Get a human review before the app handles identity documents, payments or health data. Both Crissy Field and Lovable recommend it.
What this means if you build apps with AI
The useful lesson is not "Lovable is unsafe". It is that where your data sits, and who is allowed to talk to it, is an architecture decision. AI app builders make that decision for you, so it is worth knowing which one your tool made.
There are two broad designs:
| Question | Browser talks to the database (Supabase pattern) | Database behind the app's own server |
|---|---|---|
| Where is the database key? | A public key ships in the page | Only in server code, never in the page |
| What protects the rows? | RLS policies on every table | The app's server routes and their auth checks |
| What does one missing rule expose? | The whole table, to anyone with the public key | Whatever the unprotected route returns |
| Can a scanner find many apps with one pattern? | Yes, the study's core finding | Harder, there is no public database key to reuse |
| What can still go wrong? | Missing or loose policies | Routes that skip authorization checks |
Neither design is automatically safe. The second one removes the exact failure mode in this study, but a server route that forgets to check who is asking can still leak data.
Totalum is built on the second design. Per its data storage and security docs, every project is a Next.js app running in its own Cloudflare Worker with its own keys, and each project gets its own SQL databases (one for development, one for production) that are never shared with another project. Sign-in uses BetterAuth; generated apps hash passwords, lock out repeated failed sign-ins and check sign-ups with a captcha. Every agent run takes a restore point, and you can restore the database to any minute of the last 30 days.
We ran a small first-hand check on November 15, 2026. Our own marketing site at www.totalum.app is itself a Totalum project, so we downloaded its HTML and all 22 JavaScript bundles it loads (667 KB) and searched them for Supabase keys, service-role keys, JWTs, database URLs and secret API keys. There were zero hits. The only matches for "supabase" and "tlm_sk_" were marketing copy and a masked example string. That is one site, not a study, but it is the same check Crissy Field ran, and the design means there is no database key in the page to find.
What Totalum does not do is review your authorization logic for you. If you ask for an admin page, ask for it to be admin-only, then log in as a normal user and confirm you are blocked. The same "test logged out" habit applies on every builder. For a wider look at which tools hold up once real users arrive, see our comparison of production-ready AI app builders, and if Supabase itself is the part you want to leave, our Supabase alternative guide walks through the trade-offs.
Building an app that will hold customer data? Start free on Totalum and get a Next.js app with its own database, sign-in and backups, where the database key never reaches the browser.
Lovable security vs other AI app builders: the honest summary
Lovable has better security tooling than it had in 2025, and the study does not show a breach of Lovable's own systems. It shows that a browser-to-database design puts the last line of defense in the hands of the app owner, and that many owners do not know the rule exists. If you stay on Lovable, run the scans and read the policies. If you are choosing a builder for an app that will store personal data, ask where the database key lives before you ask anything else. Our Lovable vs Totalum comparison covers the other differences, from pricing to code ownership.
FAQ
Was Lovable hacked in November 2026?
No. The Crissy Field study did not break into Lovable or Supabase. It read data that individual Lovable apps left readable because their database access rules (Row Level Security) were missing or too permissive.
How many Lovable apps exposed data?
Of 24,664 Supabase-backed Lovable sites that sent database records on first load, 7,403 (30%) had tables containing email addresses that could be read without login. The researchers say this is not a confirmed leak rate, and their sample of 159,041 projects is not representative of all Lovable apps.
Is Lovable safe to use?
Lovable can be used safely, but security depends on how each app is configured. Run the built-in Deep scan, enable RLS with scoped policies on every Supabase table, test the app while logged out, and get a professional review before handling sensitive data.
What is Row Level Security in Supabase?
Row Level Security is a Postgres feature that Supabase uses to decide which rows a request may read or change. Because Supabase apps send a public key to the browser, RLS policies are the main protection for the data.
Do other AI app builders have the same problem?
Any app that lets the browser query the database with a public key depends on the same kind of rules. UpGuard found readable tables on 16,326 of about 300,000 Supabase-linked domains, according to Trending Topics. Builders that keep the database behind the app's own server avoid this pattern, but still need correct authorization checks in their routes.
What was the April 2026 Lovable incident?
Between February 3 and April 20, 2026, a backend regression let any logged-in Lovable user read the chat history and source code of public projects. Lovable fixed it within two hours of the public report and published a post-mortem on April 22, 2026.
Conclusion: check your rules, then pick your architecture
The November 2026 Lovable security study is a warning about a design, not just about one company. When the browser can query your database, every table needs a correct rule, and one forgotten waitlist table is enough to leak. Check your app today with the steps above. For your next build, choose a setup where the database key stays on the server.
Ready to build an app your users can trust with their data? Start free at totalum.app and describe what you want to build.